To answer the question, we have to trace the root of programming language: why John Backus, Grace Hopper et al envisioned the compiler to generate machine code via pure reasoning, so that human/LLM can focus on the problem framing and design intent. https://t.co/4CgCh8cwcY
Assembly is a programming language. It is taught is schools. It is used to produce industrial-strength software.
Why can’t AI use assembly?
It can.
Now once you accept assembly, why not binary?
Comments, label/function names, source-level structure, assembler directives…. will be missing. So what?
@valigo Even an algebric datatype would help enormously in golang. I wish I could send 8/16/32-bit arrays over the same channel, like an std::variant<u16[], f32[]>.
@KirillGoldinBiz@GergelyOrosz The strategy of accelerated code rot is often rebranded as the "let it burn protocol". But, it only works if the organization does not make a person accountable/liable for the failing legacy system. Incentives matters.
@christophcsmith Like this? The insider joke that Einstein is bad at arithmetic and his distaste on probabilistic approach to quantum physics? https://t.co/JFN7YXAo74
Having programmed the Snapdragon 888 (aka Kona) myself, I am not so sure. It might end up similar to the fate of the Merlin compiler; it got absorbed into the Xilinx ecosystem after acquisition.
Qualcomm just paid $3.9 billion for a 150-person company. The prize: a programming language that lets AI run on ANY chip without NVIDIA's CUDA.
Meta and Microsoft are already placing orders.
The NVIDIA software monopoly just got its first real challenger. $3.9B says this is real.
@lauriewired Even today, I thank the gamers for absorbing the upfront cost of gaming GPUs and custom liquid cooling, so that budget-limited researchers never have to pay full price. Similar to the second hand car market.
@EzzeddineAlSai@sytelus I suppose the author quantifies efficiency in joules: energy spent to move digits in and out of the GPU transistors. By this definition, LLM sounds really brute-force to answer simple math questions, e.g. proving that sqrt(5) is an irrational number.
As an ex-researcher, WebUI has been a blessing for us who cannot afford Labview to build an instrument control panel. Most of us started out streaming Matplotlib plots in Flask/JQuery, and then slowly migrate to Vue3 with dynamic SVG/Canvas. Whoever invented MVVM is a genius.
Interesting discussions about whether browsers (and, by extension, things like Electron or Tauri) are fast, slow, bloated, or genuis.
It's actually all of this together. All of it is true. And the reason is - DOM model.
You see, DOM, the way it's designed, just physically can not be fast. The design inherently goes against the way computers work, it's just unoptimizable. Mainly because it wasn't designed for kinds of workloads we throw at it today.
With that said, it's absolutely miraculous, or, rather, genius, what engineers were able to do to optimize browsers and JavaScript engines to the extent they did.
You can definitely both applaud it, and say that time of some biggest brains in the industry was wasted, and could be spent on something better than optimizing this poorly aged design.
So I hope it makes more sense now, how same people can call browsers an engineering miracle, and then call them a bloated mess in the next sentence.
I had a similar existential crisis as I read the AOSP camera buffer and SurfaceFlinger code. There are at least 4 layers of abstractions (Java > JNI > c++ > CamX) before reaching the real ISP stuff. Makes me miss @marclevoy 's Frankencamera back in the good old days.
Modern programming scares me in many respects, where they will just build layer after layer after layer that does nothing except translate. It confuses me to read a program which you must read top-down. It says “do something.” And you go find “something.” And you read it and it says, “do something else” and you go find something and it says, “do something else” and it goes back to the top maybe. And nothing gets done. It's just relegating the problem to a deeper and deeper level. I can't keep it in my mind—I can't understand it.
Ken Thompson, 2009, in Coders At Work.
@nikstankovic_ To learn authentic Chinese, you have to read the minds of struggling immigrant kids learning English in the US/UK schools. Basically to "unlearn" all prepositions, subordinating conjunctions, phrasal verbs, gerunds, and part of speech. https://t.co/OaxeCn9duy
@Truth__Enjoyer@dalmaer Also note how the inverted layout enabled us to roll our eyes. I suspect that the cornea of octopus is just a transparent, immobile skin patch.
@mayukh_panja I can point out another cultural difference: startups require Facta non Verba talents. The solution is set long before the founders created the company. Head counts are created for pure execution. There is nothing to solve intellectually for late hires.